基于ELM和SVM的卫星云图分类研究

发布时间：2018-04-14 23:38

本文选题：卫星云图 + 云分类　；参考：《南昌航空大学》2014年硕士论文

【摘要】：气象卫星能够对地表及云层连续地进行大范围观测，由此得到的卫星云图蕴含着丰富的气象信息。这些信息为天气预报尤其是降雨分析提供了可靠依据。可是，随着气象卫星云图数据源数量上的爆炸式增长和内容上的极大丰富，相应的处理、分析工具的研发和应用却严重滞后。传统分类算法用于遥感图像云分类时，容易造成处理规模过大、分析过程复杂以及陷入局部极小值等问题，而且在分类速度和分类精度远远无法满足需求。因此，对卫星云图进行准确、快速的自动分类一直是遥感领域众多学者和科研人员的研究热点。着眼于此，本文将一种新型的单隐层前馈神经网络算法——极限学习机（Extreme Learning Machine，ELM）应用于遥感卫星云图分类中的分类器构建。另外，本文还采用了支持向量机算法进行云分类，与极限学习机分类效果进行对比分析。本文主要内容和研究成果概述如下：（1）首先介绍了论文的选题背景和意义，然后详细介绍了云分类的研究历程和现状，并对云的分类方法进行了深入的分析。（2）介绍了气象卫星及卫星云图的概念，云的种类及其在卫星云图上的表现特性，详细讲述了本文所使用的样本文件格式及读取方法，分析了遥感云图的特性和分类理论。（3）详细研究了极限学习机的学习过程，说明了该算法在学习性能上的优势和特性，并创新性地将极限学习机算法应用于遥感卫星云图分类。基于上述实验的结果，详细分析了ELM算法中隐藏层节点数对分类结果，包括分类精度和分类时间的影响，研究了其变化的规律。（4）为了进行对比，本文利用支持向量机算法设计分类器，并对相同的分类样本进行测试，分析两种算法的优劣势。最后通过两组结果的对比可以得出，，用ELM方法进行云分类是有效且分类速度上有明显优势，但是分类精度低于SVM。
[Abstract]:The meteorological satellite can continuously observe the surface and clouds in a wide range, and the resulting satellite cloud images contain abundant meteorological information.This information provides a reliable basis for weather forecast, especially for rainfall analysis.However, with the explosive growth in the number of data sources and the abundance of the contents, the research and development and application of analytical tools are lagging behind.When the traditional classification algorithm is used in cloud classification of remote sensing images, it is easy to cause problems such as too large processing scale, complex analysis process and falling into local minima. Moreover, the speed and accuracy of classification can not meet the requirements.Therefore, accurate and fast automatic classification of satellite cloud images has been a hot spot of many researchers and researchers in remote sensing field.In this paper, a new single-hidden layer feedforward neural network algorithm, extreme Learning Machine (ELM), is used to construct a classifier for remote sensing satellite cloud image classification.In addition, the support vector machine (SVM) algorithm is used for cloud classification.The main contents and research results of this paper are summarized as follows:Firstly, the background and significance of this paper are introduced, then the research history and present situation of cloud classification are introduced in detail, and the method of cloud classification is deeply analyzed.This paper introduces the concept of meteorological satellite and satellite cloud image, the category of cloud and its characteristics on satellite cloud image, describes in detail the format of sample file and the reading method used in this paper, and analyzes the characteristics and classification theory of remote sensing cloud image.(3) the learning process of LLM is studied in detail, and the advantages and characteristics of the algorithm in learning performance are explained, and the LLM algorithm is innovatively applied to the classification of remote sensing satellite cloud images.Based on the above experimental results, the effects of the number of hidden layer nodes in ELM algorithm on classification results, including classification accuracy and classification time, are analyzed in detail.In order to compare, the support vector machine (SVM) algorithm is used to design the classifier, and the same classification samples are tested to analyze the advantages and disadvantages of the two algorithms.Finally, through the comparison of the two groups of results, it can be concluded that the cloud classification using ELM method is effective and has obvious advantages in classification speed, but the classification accuracy is lower than that of SVM.
【学位授予单位】：南昌航空大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP751

【参考文献】